{"id":"W4386242325","doi":"10.1145/3600211.3604690","title":"Self-Destructing Models: Increasing the Costs of Harmful Dual Uses of Foundation Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Open Philanthropy Project; Canadian Institute for Advanced Research","keywords":"Computer science; Adversarial system; Artificial intelligence; Machine learning; Reinforcement learning; Foundation (evidence); Dual (grammatical number)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001238058,0.0001111264,0.0001736646,0.0001398615,0.000174301,0.00007107663,0.0006227366,0.00004620905,0.00000647464],"category_scores_gemma":[0.0002783563,0.00008484154,0.00005542017,0.0008018027,0.00006129703,0.001251481,0.0005840702,0.0001570785,0.000006339712],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005081196,"about_ca_system_score_gemma":0.00009116721,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006614287,"about_ca_topic_score_gemma":0.00001500416,"domain_scores_codex":[0.998535,0.000257795,0.0003469323,0.0002452139,0.0004084213,0.0002066822],"domain_scores_gemma":[0.9983208,0.0007817264,0.0002623966,0.0004269494,0.0001764851,0.00003160895],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000005580382,0.000009932384,0.0004110135,0.0000184568,0.0000180788,0.000001295508,0.001601961,0.6979362,0.0002563768,0.2903103,0.000009735901,0.009421075],"study_design_scores_gemma":[0.0001629504,0.00001794453,0.0002771525,0.00003155416,0.00001000845,0.00001028663,0.0004251685,0.9594194,0.0006033937,0.03894915,0.00000832623,0.00008467131],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2813092,0.000008136406,0.714519,0.0002439307,0.0001668217,0.000110309,4.876985e-7,0.0002790826,0.003363055],"genre_scores_gemma":[0.8376878,0.000008932673,0.1621985,0.00002555775,0.00004108011,0.000003340463,0.000002065463,0.00001058471,0.00002214262],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5563786,"threshold_uncertainty_score":0.3459738,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04462146224270767,"score_gpt":0.2854168640465077,"score_spread":0.2407954018038,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}